Trellis.2 and Pixal3D are now native in ComfyUI

🤖 Trellis.2 and Pixal3D are now native in ComfyUI

Two open 3D models have been built into the ComfyUI core (version 0.34.0+): Trellis.2 from Microsoft — 4 billion parameters, geometry and PBR materials from a single image, effective resolution up to 1536³ — and Pixal3D from Tsinghua University and Tencent ARC Lab (SIGGRAPH 2026), whose 3D model strictly matches the input image.

🌍 Image-to-3D no longer requires manual assembly: no custom nodes, compilable CUDA extensions, or PyTorch/CUDA downgrades are needed, and the code and weights of Trellis.2 are open for commercial use under MIT.

👤 Update ComfyUI to 0.34.0+, download the weights from the official library — and from a single image you will get a mesh with PBR materials and UV unwrapping, without custom nodes or version pinning.

Source 1: https://blog.comfy.org/p/trellis2-and-pixal3d-are-now-native Source 2: https://docs.comfy.org/tutorials/3d/trellis2